| ChatGPT For Dummies |
Page 2 of 4
Chapter 3: Writing Prompts for ChatGPT How you write your prompts (questions) is the key to getting the best responses from ChatGPT. Recently there has been a spate of jobs for Prompt Engineering, some commanding large salaries. There’s a discussion on prompt basics, where entering a basic prompt will give you a search engine or Wikipedia like answer. However, you need to use ChatGPT in a different way, you can give it a persona or role (e.g. As John Lennon, write me a new poem), and you yourself can take a on persona (e.g. I am the owner of a small bakery, write me a 5-year expansion plan). You can also tell ChatGPT to reply for a given type of audience (e.g. primary school pupils). You can create a context for your questions (e.g. I am the candidate for a school vote, how can I increase my popularity?). You can ask ChatGPT To summarize a response. ChatGPT remembers the context of your chat, so you can ask for more detail concerning its responses. When you’re finished with a given topic, you can start a new chat, with a new prompt. Asking new (unrelated) questions in an existing chat can confuse ChatGPT, and give bad answers. Similarly, asking too many questions in a chat can also lead to problems. Currently, Bing allows 10 questions per chat session – to prevent ChatGPT giving bad responses. It's possible to delete and export chats. It can be useful to think of working with ChatGPT as a conversation, where questions are asked, and depending on the response, further, more specific questions are asked. Ideally ChatGPT would ask for more clarification or detail, but it doesn’t work like that :-). The chapter ends with a very helpful list of prompt tips, including:
This chapter provides a host of useful suggestions on how to get the most from ChatGPT, following its advice is sure to give you more relevant and interesting responses. Chapter 4: Understanding GPT Models in ChatGPT ChatGPT models evolve quickly, GPT-3 was released in November 2022, GPT-3.5 in January 2023, and GPT-4 in March 2023. Later versions are more powerful, cover later information, and are more ‘suitable’ (less bias etc.). It’s not expected that GPT-5 will be released any time soon. GPT-4 is multimodal, meaning it can work with both text and image as input prompts, and has been trained on a much bigger set of information. OpenAI are also the creators of DALL-E, a powerful image generator, which also uses the same GPT models. The chapter moves on to looking at the pros and cons of using the different models. If you’re dealing with images, you’ll want to use GPT-4. There’s a useful table showing the scores of the various models taking various exams, with the later versions gaining higher marks. GPT-4 typically needs to be paid for, whereas GPT-3/3.5 are free. It is noted that GPT-4 can still hallucinate, producing incorrect information – although less than earlier models. As always, treat the output like having a junior assistant, you must check the output. Some images are of very poor quality, expanding them does not produce a clearer image (unfortunately, this occurs for some images, throughout the book). Chapter 5: Warnings, Ethics, and Responsible AI Examining AI from an ethical viewpoint is increasingly common. The ‘Responsible AI’ movement was created to ensure ethics are integrated into AI products, rather than added as an afterthought (or omitted completely). The movement’s features include:
OpenAI states it has a commitment to these principles, and has contributed to some of its frameworks. It’s noted that an AI model can be built using data and code on GitHub, for as little as $100, however these are typically very unreliable, and can deliver wrong and harmful responses. Both the USA and the EU have been considering laws to encompass the ethical use of AI. ChatGPT comes with public warnings, including privacy. Anything you enter can be used as training material. ChatGPT can be used to generate convincing but wrong output, that could be used as propaganda. Next, there’s a look at ChatGPT and copyright and Intellectual Property (IP). Any output generated by ChatGPT is not copyright protected by law. If you rework some of the words, those can have copyright applied. It’s noted that ChatGPT may have been trained on information that was copyright-protected, and its culpability is currently being examined. The chapter ends with a look at overcoming risks and liability. While later versions of ChatGPT are more reliable than earlier versions, they still need human experts to identify problems and provide safety guiderails. The use of the thumbs up/down in ChatGPT’s responses, allows you to provide feedback on the value of the responses. There’s a useful list of factors that may reduce Chat GPT’s risks and liabilities, including:
This chapter provides a useful overview of ChatGPT in relation to ethics. Again, some images cannot be read clearly. Chapter 6: Probing Professional and Other Uses for ChatGPT For the next few chapters, we look at how we can use ChatGPT to improve productivity in our daily lives and in specific types of work. ChatGPT is increasingly being integrated with existing software (e.g. Bing, Excel), with the aim of improving productivity for everyone. Remember to think about security and privacy. Microsoft, being linked to OpenAI, has a head start in integrating ChatGPT into its products, the chapter also discusses a wide range of other vendors and their products (e.g. ChatPDF). This chapter describes how ChatGPT can be used in:
In each case, an outline of how ChatGPT can help is discussed (e.g. for HR, ChatGPT can summarize a resume/CV and match it against a job description to produce a better match of candidates). |
|||||
| Last Updated ( Tuesday, 24 October 2023 ) |